WHITENET: A WHITE SPACE NETWORK FOR CAMPUS CONNECTIVITY USING SPECTRUM SENSING DESIGN PRINCIPLES Hope Mauwa, Antoine Bagula
Marco Zennaro
University of the western Cape ISAT Laboratory, Department of CS Bellville, 7535, South Africa
[email protected],
[email protected]
International Centre for Theoretical Physics T / ICT4D Laboratory Strada Costiera 11, Trieste, Italy
[email protected]
ABSTRACT To this day, the technical challenges of accessing TV white spaces through spectrum sensing can be summed up into its inability to provide maximum protection to primary users from interference. Yet, off-the-shelf spectrum sensing devices, which are emerging on the market at low cost, and the low computation and implementation complexities of the sensing technique, make them more and more attractive to the developing world. Building upon “WhiteNet”, a white space network management platform for campus connectivity, this paper proposes design principles that can be incorporated in a spectrum sensing-based white space identification system to minimise probability of causing interference to primary users. The principles are designed around the cooperative spectrum sensing model to further reduce chances of interference to primary users. Evaluation of the principles was done using real-world indoor measurements and based on a real TV transmitter-allocation at the University of the Western Cape in Cape Town, South Africa. The results reveal the relevance of using these design principles in white space networking using the emerging White-Fi protocol to boost the capacity of current Wi-Fi campus networks. Keywords— White-Fi, cooperative spectrum sensing, detection threshold, spectrum sensing principles 1. INTRODUCTION It has been widely recognized that in many regions of the developing world, poor Internet access in universities and research institutions is one of the causes of the scientific divide between developed and developing countries. In many of these regions, Wi-Fi has played a key role to connect campus communities by enabling inter-campus connectivity and access to the Internet but at lower access bandwidth compared to research institutions of the developed world. The transition from analog to digital television is a great opportunity to address this bandwidth issue in campus networks by using emerging protocols such as IEEE 802.11af, also referred Paper accepted for presentation at “Trust in the Information Society” ITU Kaleidoscope Conference, Barcelona, Spain, 9-11 December 2015, http://itu.int/go/K-2015.
to as White-Fi or Super Wi-Fi [1] to boost the current capacity of Wi-Fi networks with bandwidth acquired through secondary access to white space (WS) frequency. However, technologies and protocols have yet to mature to provide the proper WS equipment at affordable prices and WS identification, quantification and allocation techniques have yet to improved and move from the research boundaries to the implementation arena. Two main approaches of accessing unused spectrum in the TV frequency band (white spaces) for secondary use have been suggested in the literature; geo-location database and spectrum sensing. At the moment, there is a trend towards the use of only geo-location database approach in the US and Europe [2] as it guarantees high protection of the spectrum incumbents from the interference. The trend is supported by the development of the protocols such as Protocol to Access White Space (PAWS) [3] by the Internet Engineering Task Force (IETF), the IEEE 802.11af standard [4] and the IEEE 802.22 standard [5] to access spectrum database. However, in some regions of Africa, the use of a geo-location database has been questioned as the best approach to accessing TV white spaces (TVWSs) [6] due to its limitations and the abundance of TVWSs that nullify the need for stringent constraints on primary user protection. In such regions, therefore, spectrum sensing is expected to play a key role as an alternative method of accessing TVWSs. To this day, technical challenges of accessing TVWSs through spectrum sensing without causing interference to primary users have not been solved completely. In this paper, some design principles are being proposed that can be incorporated into a spectrum sensing-based WS identification system to minimise probability of causing interference to the primary users. These principles are designed around the concept of cooperative spectrum sensing. The proposed principles are: i) the use of different threshold values and ii) the deployment of virtual WSs pricing. These principles add an additional layer of protection to primary users after the cooperative spectrum sensing layer. The block diagram depicting the hierarchical flow of how the principles work is depicted in Figure 1. The rest of the paper is structured as follows: Section 2 gives a background to some of the challenges of spectrum sensing
mary user’s signal path as it propagates through the wireless medium. This leads to misinterpretation of measured data by the WS device where it thinks the channel is available and start to transmit, causing interference to the primary user. 3. PROPOSED PRINCIPLES
Figure 1. Hierarchical flow of how the principles work as a method for identifying TVWSs; Section 3 introduces the principles and discusses how they foster protection of primary users; Section 4 discusses some major existing principles that can be included in a spectrum sensing-based WS identification system; Section 5 discusses how the proposed principles can be implemented; Section 6 is a discussion of the experimental evaluation of the principles and Section 7 concludes the paper. 2. BACKGROUND INFORMATION There are several spectrum-sensing methodologies available but the most commonly used in WS identification is the energy detector-based sensing. Energy detector-based sensing works by measuring the energy contained in a spectrum band and comparing it with a set threshold value [7, 8]. If the energy level is above the threshold value, then the signal is considered present otherwise the spectrum band is considered vacant. This technique reigns superior over the other spectrum sensing techniques because of several factors: i) it is simple as it has low computational and implementation complexities [9, 8], ii) it has good performance [10, 11, 12] and iii) it is more generic as receivers do not need any knowledge on the primary users’ signals [7, 8]. Much as the energy detector-based sensing has these advantages over the other spectrum-sensing methodologies, it has some inherent challenges that make it less desirable as a means of accessing TVWSs, which can be summed up into inability to provide maximum protection to primary users from interference. One of its major challenge in relation to identifying TVWSs is that there is no standardized way of selecting the signal detection threshold that gives optimal performance, i.e. simultaneously giving low false positives and low false negatives. The value chosen as the detection threshold has a major impact on the performance of the spectrum sensing equipment. If the value is too high, the technique fails to detect the presence of a TV signal in a channel thereby causing harmful interference, and if the value is too low, it gives false detection when there is actually no TV signal in a channel. Another challenge of this technique is that it suffers from multi-path fading or shadowing that results into the hidden user problem [9]. In this scenario, a WS device is unable to detect the presence of a primary user service in a channel due to obstacles that block the pri-
Design paradigm underlying any suggested model based on spectrum sensing aims at eliminating its technical challenges. This section discusses the proposed principles that are being incorporated in the spectrum sensing-based WS identification component of WhiteNet; a white space networking platform under development at the University of the Western Cape (UWC) in South Africa with the expectation of resolving some of the technical challenges associated with this method of identifying WSs. 3.1. Using more than one detection threshold Deciding on the threshold to be used in spectrum sensing is a challenging issue that has been at the heart of debates concerning an absolute value to be used. To get around this problem, we are proposing to use more than one threshold value to compromise the two extremes, many false negatives or many false positives, the likely results when a single threshold is used. Measurement studies have shown that the sensitivity threshold of -114 dBm for Advanced Television Systems Committee (ATSC) TV signal detection as mandated by the Federal Communications Commission (FCC) is too conservative [13, 14, 15, 16]. -114 dBm is said to be conservative because it leads to significant loss of WSs [16]. Some studies have confirmed that, for example, [13] found no TVWSs in all the locations where the studies were done in China when a sensing sensitivity threshold of -114 dBm was used. However, relaying on the analog terrestrial television (ATT) database as ground-truth data for the ATT channel occupancy situation in Beijing, setting the sensitivity threshold to -97 dBm was enough to find WS ATT channels in indoor scenarios. On the other hand, different signal detection thresholds have been used by different studies to find WS. Therefore, using more than one threshold in the range from -114 dBm to a value that is dependent on a country’s TV broadcasting allocation scheme for transmitting sites seems to be a logical solution and is being proposed here. The FCC’s mandated detection threshold of -114 dBm is being proposed as the start threshold because it is conservative and also able to find WSs in some environments although it ends up with no WSs in others. Identifying WSs in this way helps to group WSs based on the threshold values used to detect them. If there is a request for WS use from WS devices, allocation starts with WSs detected with the lowest detection threshold and if they are not enough to satisfy the demand, then the next slot of WSs identified using the next higher detection threshold is used and so on. Based on the assumption that at each point in time, the demand for white space use from white space devices is satisfied well before using white
spaces identified with higher threshold values, the approach of identifying WSs using different thresholds and starting the allocation with WSs identified with the lowest thresholds minimizes the chances of interference to primary users due to false negatives than using random or haphazard allocation of WSs identified with a single threshold. WSs identified with higher thresholds are the most likely thresholds that may result into interference. The approach also solved the problem of resulting with either too many false negatives or too many false positives when one threshold is used to identify the TVWSs. 3.2. Virtual pricing of white spaces Another principle being proposed in this work to minimize interference to primary users from WS devices is to virtualprice WS channels within each group based on some common quantity associated with all WSs. For example, a virtual price can be given to each WS channel based on the signal strength detected in each channel with the highest price given to a channel with strongest signal and the lowest price given to a channel with the weakest signal within each group. As mentioned in subsection 3.1, the groups of WSs are based on the signal detection thresholds used to identify them. When WS devices submit requests for WS use, the cheapest WS channels within each group are allocated first. In this way, the probability of a WS device causing interference to primary incumbents if there is any false negative within the group is minimized since channels that may result into false negatives have stronger signals than channels that are actually WSs, and as such, their virtual prices are higher than the channels that are actually WSs. Consequently, they cannot be allocated to any WS device unless all the channels that are actually WSs in that WS channel group are exhausted. 4. EXISTING SPECTRUM SENSING DESIGN METHODS This section discusses existing spectrum sensing design methods that this work considers relevant to the implementation of a spectrum sensing-based WS identification system. 4.1. Cooperative spectrum sensing Cooperation among sensing equipment is vital for the optimal performance of spectrum sensing when used as a method of identifying white spaces because a network of spectrum sensors sharing sensing information obtained from their individual locations with each other has a better chance of detecting the primary user compared to local spectrum sensing [9] by a single spectrum sensor. It is due to this reason why cooperation between sensing equipment is proposed in the literature as the solution to the hidden user problem [14, 15, 17, 18, 19] that may arise due to multi-path fading or shadowing. As mentioned in the introduction, our proposed principles rely on the results generated from cooperative spectrum sensing as the first step to minimising chances
of interference to primary users. If there is a hidden user problem after cooperative spectrum sensing, then the proposed principles help to protect further that hidden user from interference. 4.2. Channel-clustering and location-clustering As mentioned in [20] and [21], a spectrum sensing-based WS identification system must also take spectrum sensor cost as a major consideration in the design of the system as they can be expensive. To avoid random placement of the energy detectors, which could result into either waste of energy detectors, i.e., many unnecessary detectors deployed or not guarantee coverage, i.e., insufficient detectors deployed [20], it is vital to perform channel-clustering and location-clustering as proposed in [20]. Once the channel clustering and location clustering is done, the algorithm proposed in [20] can be used to determine placement positions for the energy detectors. Implementing these principles means WSs are calculated according to location clusters. Therefore, secondary users are required to identify their positions before sending a request for white space use. For detailed discussion of these principles and how they can be implemented, consult [20]. 5. ALGORITHM IMPLEMENTATION The proposed principles and the existing methods that have been discussed in this paper are not environment specific. They are general principles and methods that can be implemented in a spectrum sensing-based WS identification system meant for outdoor or indoor environment. This section shows how the proposed principles can be implemented algorithmically. 5.1. WS identification using different thresholds The method for computing TVWSs using different signal detection thresholds is presented in Algorithm 1. The algorithm shows how cooperative spectrum sensing is implemented with the principle of varying the detection threshold. The inputs to the algorithm are signal strength values of all the channels from the frequency spectrum sensors deployed and the channels under consideration. The algorithm first checks if a channel under consideration is an already identified WS using any of the previously used threshold values if any. This is done in lines 4 to 6. This helps to make sure that each channel is not identified as WS more than once as the threshold values keep changing. Once it is found that a channel is not an already WS channel, the algorithm compares the signal strength values for that channel from all the sensors deployed from line 9 to 13 to find the representative signal strength value, which is the strongest signal measured in that channel from all the sensors deployed. The strongest signal is used to calculate the relative signal strength for that channel by subtracting the current threshold from it in line 14. Then the algorithm checks if the channel is WS by checking if its relative signal strength is less than or equal to zero in line 15.
If it is found to be WS, it is added to the set of WSs for that detection threshold in line 16. The process is repeated for all the channels using the current threshold value (lines 3 to 20). Once all the channels are considered using the current threshold value, the next threshold value is considered (line 24) and the process is repeated from the beginning (from line 2). This process is repeated until all the threshold values have been considered. The output from this algorithm is the set of sets of WS channels SC identified using different thresholds and the set of sets of signal strength values SS corresponding to the set of sets of all WS channels SC.
1 2 3 4 5 6 7
8 9 10 11 12 13 14
15 16 17 18 19 20 21 22 23 24 25 26 27
5.2. Compute virtual prices of WS channels Once WSs channels have been identified using the different threshold values, Algorithm 2 follows to compute their virtual prices based on signal strength recorded in each channel. Algorithm 2: Compute virtual prices of white space channels identified input : SS = {SS(1), SS(2), SS(3), ..., SS(x)}. output: V P = {V P (1), V P (2), V P (3), ..., V P (x)}. {VP is a corresponding set of sets of virtual prices of white space channels}
Algorithm 1: Identify white space channels using different 1 initialize j ← 1, strongestSignal ← 0; thresholds 2 for i ← 1 to x do input : Two-dimensional matrix st of size m by n of signal 3 while SS(i) has elements do strength values, set 4 if strongestSignal < ss(i)(j) then CH = {ch(1), ch(2), ch(3), ..., ch(m)} of 5 strongestSignal ← ss(i)(j); channels. {m is the number of channels under 6 j ← j + 1; consideration; n is the number of sensors deployed} 7 end output: SC = {SC(1), SC(2), SC(3), ..., SC(x)}, where 8 end x is less than or equal to number of threshold 9 for a ← 1 to (j − 1) do values, SS = {SS(1), SS(2), SS(3), ..., SS(x)}. 10 vp(i)(a) = |ss(i)(a)|/|strongestSignal|; {SC is a set of sets of white space channels; SS is a 11 add vp(i)(a) to V P (i); corresponding set of sets of signal strength values 12 end of the white space channels} 13 initialize j ← 1 initialize t ← startT hreshold, x ← 1; 14 end repeat 15 return VP; for i ← 1 to m do if SC is not empty then The input to the algorithm is SS, the output from Algorithm check if ch(i) is in any of SC subsets; 1. The algorithm first searches through the set of signal end strength values SS(i) to find the strongest signal in that set if ch(i) is not found in any SC subsets or SC is in lines 2 to 7. Then algorithm calculates the virtual price of empty then each WS channel by dividing its absolute signal strength with strongestSignal ← 0; the absolute strongest signal in lines 9 to 12. The process is for j ← 1 to n do repeated for each WS channel group SS(i) using the strongest if st[i][j] > strongestSignal then signal in that group and the signal strengths of WS channels strongestSignal ← st[i][j]; in the group until all WS channel groups are considered. The end output of the algorithm is the set of sets of virtual prices VP end corresponding to the set of sets SS of signal strength values rss(i) ← strongestsignal − t // rss(i) is for the WS channels. representative relative signal strength for channel i if rss(i)
